Common Misunderstandings of Survival Time Analysis

Similar documents
Kaspar Rufibach (F. Hoffmann-La Roche) Joint work with Mouna Akacha (Novartis)

Systematic Reviews. Simon Gates 8 March 2007

Locoregional treatment Session Oral Abstract Presentation Saulo Brito Silva

Making comparisons. Previous sessions looked at how to describe a single group of subjects However, we are often interested in comparing two groups

CHL 5225 H Advanced Statistical Methods for Clinical Trials. CHL 5225 H The Language of Clinical Trials

Surveillance report Published: 17 March 2016 nice.org.uk

Real World Cost-Effectiveness of Cancer Drugs

Possible causes of difference among regions How to look at the results from MRCT Non-compliance with GCP and/or protocol Apparent Differences (Play of

MUD SCT for Paediatric AML?

The Current Status of Immune Checkpoint Inhibitors: Arvin Yang, MD PhD Oncology Global Clinical Research Bristol-Myers Squibb

Evidence-Based Medicine and Publication Bias Desmond Thompson Merck & Co.

TRANSPARENCY COMMITTE OPINION. 19 December 2007

CCSS Concept Proposal Working Group: Biostatistics and Epidemiology

Immunotherapy in the Adjuvant Setting for Melanoma: What You Need to Know

Biostatistics Primer

The availability and cost are obstacles to using pvl in monitoring HIV treatment outcomes in resource-constrained settings

Maurille Feudjo Tepie Director, Observational Research, Amgen Ltd

A Case Study of Graphics in Clinical Trials: The Role of Statistical Graphics in the Recent Submission/Approval of GSK's Votrient in the US

Dr Claire Burney, Lymphoma Clinical Fellow, Bristol Haematology and Oncology Centre, UK

Estimands for time to event endpoints in oncology and beyond

Supplementary Appendix to manuscript submitted by Trappe, R.U. et al:

DISCLOSURE SLIDE. ARGOS: research funding, scientific advisory board

Evidence Based Medicine

Lecture Outline. Biost 590: Statistical Consulting. Stages of Scientific Studies. Scientific Method

Scottish Medicines Consortium

Lecture Outline. Biost 517 Applied Biostatistics I. Purpose of Descriptive Statistics. Purpose of Descriptive Statistics

Benefit - Risk Analysis for Oncology Clinical Trials

Case Studies in Bayesian Augmented Control Design. Nathan Enas Ji Lin Eli Lilly and Company

Title: Lung cancer (non-small-cell, anaplastic lymphoma kinase positive, previously treated) ceritinib

EMA Pediatric Web Synopsis Protocol A November 2011 Final PFIZER INC.

Total 49 chromosomes. No. of cases Age Median (range) < Sex Female Male 65 (0-83) 9 (13%) 32 (48%) 26 (39%)

Epidemiologic study designs

An Overview of Survival Statistics in SEER*Stat

AN INDEPENDENT VALIDATION OF QRISK ON THE THIN DATABASE

Scottish Medicines Consortium

Background 1. Comparative effectiveness of nintedanib

First Line Management of Classical Hodgkin Lymphoma

HIV/AIDS CLINICAL CARE QUALITY MANAGEMENT CHART REVIEW CHARACTERISTICS OF PATIENTS FACTORS ASSOCIATED WITH IMPROVED IMMUNOLOGIC STATUS

Biostatistics II

Survival in Teenagers and Young. Adults with Cancer in the UK

CHILDHOOD CANCER SURVIVOR STUDY Long-Term Morbidity in Survivors of Childhood Leukemia with Down Syndrome Analysis Concept Proposal

Statistical Methods for the Evaluation of Treatment Effectiveness in the Presence of Competing Risks


Evidence tables from the systematic literature search for premature ovarian insufficiency surveillance in female CAYA cancer survivors.

Summary. Table 1 Blinatumomab administration, as per European marketing authorisation

Cumulative metformin use and its impact on survival in gastric cancer patients after INDEX

Statistical techniques to evaluate the agreement degree of medicine measurements

Supplement for: CD4 cell dynamics in untreated HIV-1 infection: overall rates, and effects of age, viral load, gender and calendar time.

CLINICAL STUDY REPORT SYNOPSIS

Retention in HIV care predicts subsequent retention and predicts survival well after the first year of care: a national study of US Veterans

Outcome of acute leukemia patients with central nervous system (CNS) involvement treated with total body or CNS irradiation before transplantation

The adolescents and young adults population (AYA) has

R. A. Nout Æ W. E. Fiets Æ H. Struikmans Æ F. R. Rosendaal Æ H. Putter Æ J. W. R. Nortier

Everolimus for the Treatment of Advanced Renal Cell Carcinoma (arcc ) - Additional Analysis Using RPSFT

Technology appraisal guidance Published: 27 July 2016 nice.org.uk/guidance/ta399

The role of liposomal doxorubicin in DLBCL lymphomas. Michele Spina Aviano

Types of Data. Systematic Reviews: Data Synthesis Professor Jodie Dodd 4/12/2014. Acknowledgements: Emily Bain Australasian Cochrane Centre

How to Conduct a Meta-Analysis

Have we moved beyond EPOCH for B-cell non-hodgkin lymphoma? YES!

Sequencing of therapies in mrcc. Ari Hakimi MD Assistant Professor Urology Service, Department of Surgery MSKCC

Identifying Geographic & Socioeconomic Disparities in Access to Care for Pediatric Cancer Patients in Texas

STATISTICAL MODELLING FOR THE PROGNOSTIC CLASSIFICATION OF PATIENTS WITH PANCREATIC CANCER FOR OPTIMISATION OF TREATMENT ALLOCATION

Accelerating Innovation in Statistical Design

Antihypertensive Trial Design ALLHAT

NATIONAL INSTITUTE FOR HEALTH AND CLINICAL EXCELLENCE

Leveraging Standards for Effective Visualization of Early Efficacy in Clinical Trial Oncology Studies

Discussant Primary and Secondary Prevention of CV Events in the CANVAS Program

EGFR inhibitors in NSCLC

Scottish Medicines Consortium

Overview. All-cause mortality for males with colon cancer and Finnish population. Relative survival

Professor Mark Bower

Introduction ORIGINAL RESEARCH

Confounding and Bias

Evaluation of AJCC, UICC, and Brigham and Women's Hospital Tumor Staging for Cutaneous Squamous Cell Carcinoma

Debate 1 Are treatments for small cell lung cancer getting better? No:

Shared Care & Survival CTYA SSCRG (Childhood Cancer Research Group)

Risk Factors for Loss to Follow-up During Active Surveillance of Patients with Stage I Seminoma

Methodology for Statistical Analysis of

Statistical Analysis Plan: Post-hoc analysis of the CALORIES trial

Statistical Considerations for Post Market Surveillance

Pacira v. FDA: Summary of Declaration by Lee Jen Wei, PhD Concluding that the Pivotal Hemorrhoidectomy Study for EXPAREL Demonstrated a Treatment

Learning Objectives 9/9/2013. Hypothesis Testing. Conflicts of Interest. Descriptive statistics: Numerical methods Measures of Central Tendency

Chest pain and subsequent consultation for coronary heart disease:

TRANSPARENCY COMMITTEE OPINION. 8 November 2006

Meta-analysis: Basic concepts and analysis

Effect of anastrozole and tamoxifen as adjuvant treatment for early-stage breast cancer: 10-year analysis of the ATAC trial

Technology appraisal guidance Published: 8 November 2017 nice.org.uk/guidance/ta487

Optimal Management of Isolated HER2+ve Brain Metastases

Munich Cancer Registry

Live WebEx meeting agenda

Cost-effectiveness of osimertinib (Tagrisso )

symposium article introduction symposium article

Understanding Clinical Trials

Synopsis (C1034T02) CNTO 95 Module 5.3 Clinical Study Report C1034T02

Biomarkers in oncology drug development

CHILDHOOD CANCER SURVIVOR STUDY ANALYSIS CONCEPT PROPOSAL

INTRODUCTION TO SURVIVAL CURVES

HIV-HBV coinfection in HIV population horizontally infected in early childhood between

UK Liver Transplant Audit

Neodjuvant chemotherapy

Transcription:

Common isunderstandings of Survival Time Analysis ilensu Shanyinde Centre for Statistics in edicine University of Oxford 2 nd April 2012

Outline Introduction Essential features of the Kaplan-eier survival curves edian survival times edian follow-up times Forest plots to present outcomes by subgroups

Introduction Survival data is common in cancer clinical trials Survival analysis methods are necessary for such data Primary endpoints are considered to be time until an event of interest occurs Some evidence to indicate inconsistencies or insufficient information in presenting survival data Poor presentation could lead to misinterpretation of the data 3

Paper by DG Altman et al (1995) Systematic review of appropriateness and presentation of survival analyses in clinical oncology journals Assessed; - Description of data analysed (length and quality of follow-up) - Graphical presentation 4

Cohort sample size 132, paper publishing survival analysis Results; - 48% did not include summary of follow-up time - edian follow-up was frequently presented (58%) but method used to compute it was rarely specified (31%) Graphical presentation; - 95% used the K- method to present survival curves - Censored observations were rarely marked (29%) - Number at risk presented (8%) 5

Paper by S athoulin-pelissier et al (2008) Systematic review of RCTs, evaluating the reporting of time to event end points in cancer trials Assessed the reporting of; - Number of events and censoring information - Number of patients at risk - Effect size by median survival time or Hazard ratio - Summarising Follow-up 6

Cohort sample size 125 Results; - Survival analysis cited with K- method (92%) - Censoring was less defined (47%) - Number of patients at risk was less defined (45%) - edian Follow-up was reported (71%), but method used was not specified in almost all the papers 7

Presenting Kaplan eier survival curves 8

Case 1: Presenting K- survival curves Patients with first relapse of acute lymphoblastic leukaemia aged 1-18 years Experiencing two types of relapse, isolated or combined relapse Cohort size N = 123, of which 80 (65%) had i-cns relapse 58 patients experienced the event (death) 9

Case 1 :- Presenting K- survival curves Assess overall survival by type of relapse (isolated and combined relapse) K- method estimates survival probability in both groups Graphical display of the survival curves using the K- method To effectively display an informative plot as recommended 10

0.00 0.25 0.50 0.75 1.00 Overall survival probability 0.00 0.25 0.50 0.75 1.00 Isolated CNS Combined CNS Isolated CNS Combined CNS Log rank p = 0.3026 Log rank p = 0.3026 0 12 24 36 48 60 Time (months) Number at risk Isolated CNS Combined CNS 0 12 24 36 48 60 Time (months) 80 53 38 25 12 10 43 28 20 12 9 4 Number at risk Isolated CNS Combined CNS 80 53 38 25 12 10 43 28 20 12 9 4 11

edian survival times 12

edian Survival time Effect size is sometimes determined using edian survival time, if incorrectly presented could mislead results edian survival time : - Time when half of the patients are event free edian survival time estimated from the K- survival curves. Takes into account patients who have been censored, so all patients are included 13

Case 2: edian Survival time - Retrospective study of patients newly diagnosed with Hodgkin Lymphoma (HL) - Cohort size N=224, data collected from 1997 2010 from five hospitals - Patients comprising of 93 (42%) HIV+ patients - 31 events (deaths) of which 15(48%) were HIV+ patients - To compare overall survival of patients according to HIV status 14

Case 2: edian Survival time Principal Investigator proposed summary of observed survival times excluding censored patients 16 HIV and 15 HIV + patients have died at a median time of 32 months (range: 5-97) and 9 months (range: 1-75) respectively, p= xx 15

0.00 0.25 0.50 0.75 1.00 Cumulative percentage 0.00 0.10 0.20 0.30 Case 2: edian Survival time 16 HIV and 15 HIV + patients have died at a median time of 32 months (range: 5-97) and 9 months (range: 1-75) respectively, p= xx Five-year overall survival (OS) was 81% (95%CI: 69-89) and 88% (95%CI:80-93) for HIV positive and negative patients respectively. HIV+ HIV- HIV+ HIV- log rank p=0.148 0 5 10 15 Time (years) Number at risk HIV+ 93 28 8 0 HIV- 131 70 28 0 0 5 10 15 Time (years) Number at risk HIV+ 93 28 8 0 HIV- 131 70 28 0 16

edian follow-up 17

edian follow-up Quantify length of follow-up of patients The median follow-up is an indicator of how mature your survival data is (e.g. how many months on average the patients were followed since randomisation into the study). Interpretation depends greatly on the time frame in which the study was carried out i.e. did we observe enough events Several methods (yielding different results) could be used and need to report method used in analysis 18

ethod 1: edian follow-up ethod 1 - edian follow-up using all patients - Have data on date of event or date patient last seen - Estimated from observed follow-up times - Advantages; - Includes all patients Disadvantages - Unstable and biased towards patients with short follow-up - Directly affected by times of observed events 19

ethod 2: edian follow-up ethod 2 - edian follow-up for censored patients or survivors - Estimated from observed follow-up times, excluding patients with events - Advantages; - We are not aware of how long we could have followed the patient if they had not experienced the event - Disadvantages; - Loss of information - Unstable estimate when number of survivors is small 20

ethod 3: edian follow-up ethod 3 - Reverse K- method - Estimated from the from K- method, but events are reversed - The event of interest here becomes being alive and death is censored Advantages; - Analogous to the K- estimator - Robust Disadvantages; - Challenging to understand 21

Case 3: edian follow-up Patients with first relapse of acute lymphoblastic leukaemia. Two types of relapse, isolated or combined relapse Cohort size N = 123, of which 80 (65%) had i-cns relapse To summarise median follow-up for the whole cohort 22

Case 3: edian follow-up ethod All patients Censored patients only edian follow-up estimate 33.9 months 36.3 months Reverse K- 39.5 months (95%CI; 36.0-48.5) Depending on which method you use different results are obtained. Consider this in your analysis. 23

Forest plots 24

Forest Plots Recently used method of displaying lots of information in small space and getting the bigger picture across groups We present Relative Risk, but could also be used to present Hazard Ratios or Odd Ratios 25

Overall survival probability 0.00 0.25 0.50 0.75 1.00 Case 4: Forest plot Patients with first relapse of acute lymphoblastic leukaemia, aged 1-18 years Some concerns that age at relapse might have a big effect on Figure 3B outcome <10 10 Log rank p = 0.0108 0 12 24 36 48 60 Time (months) Number at risk <10 10 78 57 40 28 15 10 45 24 18 9 6 4 26

27

28

Summary Some recommended essential features when graphically displaying the survival curves edian survival times and what it actually means Report method used to obtain median follow-up Graphical display of subgroups using Forest Plots to present outcomes 29

References DG Altman et al (1995), Review of survival analyses published in cancer journals. British Journal of Cancer 72:511-518 S athoulin-pelissier et al (2008) Survival End Point Reporting in Randomized Cancer Clinical Trials: A Review of ajor Journals. J Clin Oncol 26:3721-3726 I Zweiner et al (2011), Survival Analysis. edicine 108:163-169 Schemper and TL Smith (1996), A Note on Quantifying Followup Studies of Failure Time. Controlled Clinical Trials 17: 343-346 30

Acknowledgements Sharon Love Senior Statistician Centre for Statistics in edicine, University of Oxford Dr Saha Vaskar CRUK Professor of Paediatric Oncology Academic Unit of Paediatric and Adolescent Oncology, anchester Dr Silvia ontoto Clinical Senior Lecturer/ Honorary Consultant Barts Cancer Institute- a CR-UK Centre of Excellence, Queen ary University 31